Artificial intelligence is already shaping social impact work. It influences which services people find online, how enquiries are triaged, and how funding bids, reports and recommendations are written.
For many organisations, AI promises speed and scale in systems already under pressure. But it also raises real concern.
Most social impact leaders do not want technology that embeds bias, deepens inequality or removes agency from communities. At the same time, few can afford to ignore tools that claim to improve efficiency and reach.
These tensions sit at the centre of Kara Bombell’s work at EthicAi, where she focuses on the ethical design and use of AI in systems that affect people directly. From her perspective, the question is no longer whether AI will be used. It is how it is designed, governed and held to account.
Where AI is being used
Recent global conversations on AI reveal a gap between ambition and reality. This was especially clear at the Global AI Impact Summit in Delhi, where Kara Bombell attended as part of the Australian government delegation.
Beyond keynote speeches and policy announcements, Bombell heard from practitioners working with AI in everyday social impact settings. Education systems with unreliable internet access. Welfare and social services where mistakes carry serious consequences. Health services supporting communities across vast and unevenly resourced regions.
The strongest examples shared a common starting point. They began with the problem, not the technology.
In practice, that looked like:
- Education, where teams asked how to support students with limited digital access before introducing AI‑enabled learning tools
- Social services, where organisations focused on identifying people most at risk of missing out when face‑to‑face contact was limited, before automating triage or referrals
- Health, where AI‑supported systems were designed to complement local workers rather than replace human presence altogether
These teams could clearly explain who might be excluded, which decisions carried the greatest risk, and where human judgement needed to stay firmly in the loop. Weaker examples tended to start with the tool and work backwards, searching for a problem it might solve.
That distinction matters. Many AI systems assume stable connectivity, a fixed address or a certain level of digital literacy. When those assumptions do not hold, harm is not abstract. It shows up as missed services, incorrect decisions and declining trust. For organisations focused on equity, starting with the problem is not a design preference. It is an ethical requirement.
Why ethical AI depends on people
Through its collaboration with MIT Sloan, EthicAi has focused on the human capabilities required when AI becomes part of decision making systems. This work led to the EPOCH framework.
- Empathy matters because AI can generate supportive language but cannot understand lived experience or cultural context.
- Presence matters because people notice hesitation, uncertainty and discomfort in ways systems do not.
- Opinion matters because ethical judgement and critical thinking become more important as automation increases, not less.
- Creativity matters because AI works within existing frames, while people are the ones who recognise when those frames no longer serve the outcome.
- Hope and vision matter because progress in complex social systems requires intent and imagination, not optimisation alone.
Ethical AI is not only a technical challenge. It is a human capability challenge.
Accountability does not sit with the system
One principle is clear throughout our work in adopting and implementing this technology. AI does not carry responsibility. Organisations do.
Accountability sits with those who choose to adopt AI, those who design or procure systems, and those who act on automated outputs. This includes decisions influenced by recommendations, rankings or risk scores.
For social impact leaders, responsible use means asking direct questions. Do we understand the problem before introducing technology? Do teams have the skills to interpret and challenge AI outputs? Can communities question or appeal automated decisions? Are we paying attention to where data, influence and power are accumulating?
Responsible AI is not about avoiding risk altogether. It is about being clear on where responsibility sits when things go wrong.
Why this matters now
AI adoption is accelerating across government, philanthropy and the for-purpose sector. At the same time, trust in institutions remains fragile and expectations around accountability continue to rise.
The implication is straightforward: if AI is going to support social impact rather than undermine it, human judgement must remain central. Technology can assist decision making, but it cannot replace care, responsibility or ethical choice.
Used well, AI can strengthen impact. Used poorly, it can scale harm quickly. The difference is rarely the tool itself. It is the decisions made around it.
Self‑service AI can help, but real progress often comes from structured tools, systems and specialist support. ImpactInstitute works with organisations to accelerate impact strategy, measurement, evaluation and communication, while keeping human judgement and context at the centre. Contact us to learn more.